Gemini models, with their strong general capabilities in multimodal and long-context reasoning, offer exciting possibilities in medicine. Med-Gemini, a family of highly capable multimodal models specialized in medicine, integrates web search and can be efficiently tailored to novel modalities using custom encoders. Evaluated on 14 medical benchmarks, Med-Gemini achieves state-of-the-art (SoTA) performance on 10 of them, surpassing GPT-4 on every benchmark where a direct comparison is viable. On the MedQA (USMLE) benchmark, Med-Gemini achieves a SoTA performance of 91.1% accuracy, outperforming prior models by 4.6%. Med-Gemini also excels in multimodal and long-context capabilities, achieving SoTA performance on tasks like needle-in-a-haystack retrieval from long de-identified health records and medical video question answering. It surpasses human experts in tasks such as medical text summarization and referral letter generation, demonstrating real-world utility. Med-Gemini's performance suggests real-world utility in areas like medical research, education, and dialogue. The models are built on Gemini's foundational capabilities in language, multimodal understanding, and long-context reasoning, with enhancements through self-training, web search integration, and customized encoders. Med-Gemini's long-context capabilities enable it to analyze complex medical data, including de-identified electronic health records and videos. The models are evaluated on various benchmarks, including text-based reasoning, multimodal tasks, and long-context processing, demonstrating their effectiveness in medical applications. The results highlight the potential of Med-Gemini in medicine, although further rigorous evaluation is needed before real-world deployment.Gemini models, with their strong general capabilities in multimodal and long-context reasoning, offer exciting possibilities in medicine. Med-Gemini, a family of highly capable multimodal models specialized in medicine, integrates web search and can be efficiently tailored to novel modalities using custom encoders. Evaluated on 14 medical benchmarks, Med-Gemini achieves state-of-the-art (SoTA) performance on 10 of them, surpassing GPT-4 on every benchmark where a direct comparison is viable. On the MedQA (USMLE) benchmark, Med-Gemini achieves a SoTA performance of 91.1% accuracy, outperforming prior models by 4.6%. Med-Gemini also excels in multimodal and long-context capabilities, achieving SoTA performance on tasks like needle-in-a-haystack retrieval from long de-identified health records and medical video question answering. It surpasses human experts in tasks such as medical text summarization and referral letter generation, demonstrating real-world utility. Med-Gemini's performance suggests real-world utility in areas like medical research, education, and dialogue. The models are built on Gemini's foundational capabilities in language, multimodal understanding, and long-context reasoning, with enhancements through self-training, web search integration, and customized encoders. Med-Gemini's long-context capabilities enable it to analyze complex medical data, including de-identified electronic health records and videos. The models are evaluated on various benchmarks, including text-based reasoning, multimodal tasks, and long-context processing, demonstrating their effectiveness in medical applications. The results highlight the potential of Med-Gemini in medicine, although further rigorous evaluation is needed before real-world deployment.